Blind Wave Field Characterization from Fluid Lensing
2
Blind Wave Field Characterization from Fluid Lensing EE368 Final Project Proposal, Ved Chirayath, Oscar Galvan-Lopez, Ronnie Instrella [email protected], [email protected], [email protected] Abstract Fluid Lensing is a theoretical model (Chirayath et al ) for fluid-optical interactions in turbulent flows as 1 well as two-fluid surface boundaries that, when coupled with computational imaging and an unique image- processing pipeline, may be used to not only remove strong optical distortions along the line of sight, but also significantly enhance the angular resolution of an otherwise underpowered optical system. Fluid Lensing has recently been used to map coral reefs underwater from an aerial UAV-platform at sub-cm scale for the first 23 time and enabled 3D reconstruction (Figure 1), which has proved a ground-breaking tool in modern marine research and for marine environment conservation. In our final project, we explore a novel, unsolved, problem in ocean optics - the inverse problem of Fluid Lensing. Namely, from a raw high-frame-rate video sequence, rather than just produce a distortion free image from Fluid Lensing, is it possible to uniquely characterize the wave field at an instance in time? If so, this information can not only provide an order of magnitude improvement to future Fluid Lensing implementations, but also a powerful remote sensing ability to detect features betraying ocean circulation patterns, flow properties, wave breaking and even the presence of underwater submersibles and marine fauna . We propose 4 using templates of surface wave fields coupled with optical simulations to match a measured optical distortion, as identified and tracked by computer vision algorithms, to a library of wave distortion templates and determine surface wave shape, amplitude, wavelength and group velocity uniquely. Project Summary The main purpose of this project is to uniquely determine simulated ocean surface wave conditions exclusively by using image processing algorithms on high frame rate video data. Fluid Lensing has be shown to successfully image targets underwater (Figure 1) using image processing tools and statistical models of the fluid surface, but our project aims to resolve the inverse problem and uniquely solve the fluid surface condition rather than relying on statistical models of wave properties. Such information could be used to provide an order of magnitude improvement to future Fluid Lensing implementations as well as a powerful remote sensing ability to measure ocean surface roughness. Project Methodology We propose using templates of surface wave fields coupled with optical simulations to match a measured optical distortion, as identified and tracked by computer vision algorithms, to a library of Figure 1 - Fluid Lensing results from UAV-based mapping of coral reefs
Transcript of Blind Wave Field Characterization from Fluid Lensing
Blind Wave Field Characterization from Fluid Lensing!EE368 Final Project Proposal, !!
Ved Chirayath, Oscar Galvan-Lopez, Ronnie [email protected], [email protected], [email protected]!!
Abstract!! Fluid Lensing is a theoretical model (Chirayath et al ) for fluid-optical interactions in turbulent flows as 1
well as two-fluid surface boundaries that, when coupled with computational imaging and an unique image-processing pipeline, may be used to not only remove strong optical distortions along the line of sight, but also significantly enhance the angular resolution of an otherwise underpowered optical system. Fluid Lensing has recently been used to map coral reefs underwater from an aerial UAV-platform at sub-cm scale for the first 2 3
time and enabled 3D reconstruction (Figure 1), which has proved a ground-breaking tool in modern marine research and for marine environment conservation.!! In our final project, we explore a novel, unsolved, problem in ocean optics - the inverse problem of Fluid Lensing. Namely, from a raw high-frame-rate video sequence, rather than just produce a distortion free image from Fluid Lensing, is it possible to uniquely characterize the wave field at an instance in time? If so, this information can not only provide an order of magnitude improvement to future Fluid Lensing implementations, but also a powerful remote sensing ability to detect features betraying ocean circulation patterns, flow properties, wave breaking and even the presence of underwater submersibles and marine fauna . We propose 4
using templates of surface wave fields coupled with optical simulations to match a measured optical distortion, as identified and tracked by computer vision algorithms, to a library of wave distortion templates and determine surface wave shape, amplitude, wavelength and group velocity uniquely. !!Project Summary!! The main purpose of this project is to uniquely determine simulated ocean surface wave conditions exclusively by using image processing algorithms on high frame rate video data. Fluid Lensing has be shown to successfully image targets underwater (Figure 1) using image processing tools and statistical models of the fluid surface, but our project aims to resolve the inverse problem and uniquely solve the fluid surface condition rather than relying on statistical models of wave properties.
Such information could be used to provide an order of magnitude improvement to future Fluid Lensing implementations as well as a powerful remote sensing abil i ty to measure ocean surface roughness.!!Project Methodology!We propose using templates of surface wave fields coupled with optical simulations to match a measured optical distortion, as identified and tracked by computer vision algorithms, to a library of
Figure 1 - Fluid Lensing results from UAV-based mapping of coral reefs
wave distortion templates and determine surface wave shape, amplitude, wavelength and group velocity uniquely. In Figure 2, we show a simple experimental demonstration of Fluid Lensing resolving a resolution test target.!!The test target is imaged from the bottom of a 12ft deep pool, while attached to a pole 7 feet above the surface (the inverse problem looking down is similar). A raw frame captured from a high-frame-rate video is pictured above and shows the
significant optical distortion from waves in the pool interfering in the optical path. The middle image shows the result of passing one second of data (120 frames) to the Fluid Lensing scheme to reconstruct the target accurately, albeit with reduced contrast. Once resolved, we propose using this image as a reference for comparison to further raw image data showing distortions. However, further distortions will be tracked in an attempt to characterize the wave field responsible for the optical distortions. We hope to achieve the third image; namely, a unique picture of the wave field at an instance in time just from the steps mentioned above.!!Project Objectives!This project is broken into multiple parts. (1) First, we hope to create a series of scripts that will generate a library of templates of known waves and their perturbations using videos of surface waves. (2) Using the generated library, we plan to use techniques such as template matching and match filtering to characterize ocean waves captured from aerial footage. (3) The generated wave parameters will be used to reconstruct an image of the seafloor, and the resulting reconstruction will be compared to the known structure of the floor surface. Comparisons between these images will be used to determine the accuracy of our method.!!Project Outcomes!Overall, the significance of this project is its potential to decrease the computational costs associated with image reconstruction using Fluid Lensing as well as offering a novel remote sensing capability for existing airborne and space borne instruments. If the wave library is large enough to accurately determine wave parameters from recorded video, then it may be possible to use direct optical coupling and raytracing to provide an order of magnitude increase in Fluid Lensing performance and resolution.!!References!1. www.vedphoto.com/fluid-lensing!2. Tamaki Bieri, Ved Chirayath, Trent Lukaczyk. Reactive Reefs Project. www.vedphoto.com/reactive-reefs !3. “Stanford drones open way to new world of coral research.” Stanford University News. http://news.stanford.edu/news/2013/october/coral-reefs-drones-101613.html!4. Alpers, W., and I. Hennings (1984), A theory of the imaging mechanism of underwater bottom topography by real and synthetic aperture radar, J. Geophys. Res., 89(C6), 10529–10546, doi:10.1029/JC089iC06p10529.
Figure 2 - RAW frame showing distortion (above image), middle frame resolved after
Fluid Lensing (middle image), hypothetically computed wave surface determined from using middle figure as baseline and more raw data for
measurement of optical distortions (bottom image)
